Power grid technical transformation pricing system based on industrial data analysis

Abstract

This invention relates to the field of power grid engineering cost management and industrial data analysis technology, specifically a power grid technical upgrade pricing system based on industrial data analysis. The system includes: an industrial data acquisition module, which synchronously acquires project type, timestamp, BIM structure data, IoT device status time-series data, and financial text data; a data modality alignment module, which performs cross-modal feature alignment and joint representation; a concept drift perception module, which monitors changes in feature distribution and quantifies the decay of pricing rules; a topology reconfiguration decision module, which, in conjunction with audit compliance logic, determines whether to generate a pricing topology reconfiguration instruction; a dynamic compensation generation module, which generates pricing compensation parameters and calculation logic paths and updates the pricing topology structure; an adaptive evolution module, which evaluates the interpretability index of dynamic compensation and identifies the model evolution state; and a pricing execution and output module, which outputs the final pricing result and cost list. This invention achieves dynamic, adaptive, and auditable pricing calculation for power grid technical upgrade projects.

Description

Technical Field

[0001] This invention relates to the field of power grid engineering cost management and industrial data analysis technology, specifically to a power grid technical renovation pricing system based on industrial data analysis. Background Technology

[0002] As an important part of power system construction and operation and maintenance, power grid technical renovation projects often involve a variety of factors such as equipment updates, bay expansion, construction organization adjustments, changes in material procurement, and fluctuations in the site environment. To ensure the accuracy, timeliness, and auditability of the cost calculation for technical renovation projects, it is usually necessary to comprehensively analyze the structural data, equipment status data, and cost document information generated during the project implementation period, and based on this, complete the pricing rule call, cost calculation, and cost list output. However, when calculating the cost of power grid upgrade projects, traditional methods often rely on pre-defined static quota logic and fixed pricing paths. When faced with heterogeneous data from multiple sources, such as building information model structure data, IoT device status time-series data, and unstructured financial texts, problems such as inconsistent data formats, inconsistent time granularity, and delayed updates of cost basis easily arise. Existing pricing methods usually calculate costs directly according to established rules, lacking the ability to effectively identify changes in data distribution, decreased applicability of rules, and the need to adjust pricing logic during the project process. This leads to pricing results that easily deviate from the actual cost composition when transportation conditions change, construction efficiency fluctuates, or supplementary agreements are added. This static pricing method not only affects the accuracy of cost calculation for power grid upgrade projects but may also reduce the interpretability and audit compliance of the pricing process. Summary of the Invention

[0003] The purpose of this invention is to provide a power grid upgrade pricing system based on industrial data analysis, and to solve the following technical problems: To avoid relying on static rules in the pricing system, dynamic, adaptive, and auditable pricing calculations for power grid upgrade projects are achieved under the condition of continuous changes in multi-source industrial data. The abstract changes brought about by rule decay can be transformed into specific executable compensation parameters and traceable calculation logic paths to achieve controllable reconfiguration. Furthermore, the addition of logic is evaluated through quantitative methods to achieve real-time constraints on the auditability of the reconfiguration results. Finally, the verified and safe pricing evolution results are precipitated into a transferable core structure, thereby realizing the long-term adaptive capability of the system under different power grid upgrade scenarios.

[0004] The objective of this invention can be achieved through the following technical solutions: The power grid technical upgrade pricing system based on industrial data analysis includes: an industrial data acquisition module, used to acquire the current technical upgrade project type and timestamp information, and to collect multi-source heterogeneous data to obtain raw industrial data, as well as the current pricing rules; The data modality alignment module is used to perform cross-modal feature alignment and joint representation processing on the raw industrial data, and extract basic feature representations. The concept drift perception module is used to monitor the distribution of the basic feature representation by combining a preset historical benchmark to obtain the feature distribution offset, identify the concept drift state, and quantify the rule decay degree of the current pricing rule. The topology reconfiguration decision module is used to retrieve historical pricing topology structures and predefined audit compliance logic to determine the target reconfiguration strategy, compare the rule decay degree with the preset decay threshold, and determine whether to generate a pricing topology reconfiguration instruction. The dynamic compensation generation module is used to, if the pricing topology reconstruction instruction is not generated, use the historical pricing topology as the updated pricing topology and set default pricing compensation parameters; if the pricing topology reconstruction instruction is generated, combine the target reconstruction strategy to generate pricing compensation parameters and calculation logic path, and update the pricing topology to obtain the updated pricing topology. An adaptive evolution module is used to evaluate the updated pricing topology to obtain a dynamic compensation interpretability index and identify the evolution state of the pricing model. The pricing execution and output module performs pricing calculations on the original industrial data based on the updated pricing topology and the pricing compensation parameters, and outputs the final pricing results and cost list.

[0005] Optionally, methods for obtaining raw industrial data include: Real-time acquisition of multi-channel business data, including building information model structure data, IoT device status time-series data, and unstructured financial text data; Each type of multi-channel service data is subjected to format standardization processing to obtain standard format data corresponding to each type of multi-channel service data; Based on the timestamp information, timestamp synchronization compensation is performed on the standard format data to obtain the original industrial data corresponding to each type of multi-channel business data.

[0006] Optionally, methods for performing cross-modal feature alignment and joint characterization processing on the raw industrial data include: Independent feature extraction is performed on each type of raw industrial data to obtain an independent modal feature vector; The timestamp information and the current technical renovation project type are obtained, and based on the timestamp information and the current technical renovation project type, the corresponding spatial mapping matrix is ​​obtained from the pre-constructed benchmark feature space; the spatial mapping matrix represents the degree of mutual influence between different modalities of multi-source heterogeneous data. Based on the aforementioned spatial mapping matrix, a cross-modal attention mechanism is constructed; The cross-modal attention mechanism is used to assign weights to each independent modal feature vector. Specifically, the initial attention score is processed by a normalization function to obtain the final assigned weights, and each weighted independent modal feature vector is spliced ​​and fused to complete the cross-modal feature alignment and joint representation processing of the original industrial data.

[0007] Optionally, methods for obtaining feature distribution offsets include: The basic feature representation is divided into multiple current data windows by a sliding window. Each of the current data windows is centralized to obtain centralization features; Each of the centered features is whitened to obtain a whitened feature; Principal component analysis was performed on all the whitening features to determine the main feature dimensions; From the preset historical baseline distribution, extract the baseline feature vectors corresponding to the main feature dimensions and construct the baseline matrix; The centered features of each current data window are combined to form a current feature matrix. The covariance matrix of the current feature matrix is ​​calculated. The covariance matrix is ​​decomposed to obtain the current feature vector matrix. A Gaussian radial basis function is introduced to map the current eigenvector matrix and the reference matrix to the regenerating kernel Hilbert space. The maximum mean difference between the current eigenvector matrix and the reference matrix is ​​calculated, and the maximum mean difference is used as the feature distribution offset corresponding to each current data window.

[0008] Optionally, methods for quantifying the rule decay of the current pricing rule include: The multiple feature distribution offsets are arranged in chronological order to form an offset time series, and the offset time series is smoothed to obtain a smooth offset sequence. The dynamic time warping algorithm is used to calculate the degree of matching between the smoothed offset sequence and the standard offset vectors corresponding to different attenuation levels in the pre-established regular attenuation mapping library; Extract the preset value of the attenuation level corresponding to the standard offset vector whose matching degree is greater than the preset matching threshold. The preset value represents the risk of failure of the pricing rule. The preset value is used as a candidate attenuation coefficient. If the matching degree of all standard offset vectors is less than or equal to the preset matching threshold, the current rule attenuation degree is recorded as the preset minimum default value. The largest candidate attenuation coefficient is used as the rule attenuation degree of the current pricing rule. The rule attenuation degree calculated each time is stored in the historical attenuation degree set.

[0009] Optionally, methods for determining whether to generate a pricing topology reconfiguration instruction include: A preset threshold set, which includes initial attenuation thresholds corresponding to different types of technical renovation projects; Based on the current technical upgrade project type, obtain the corresponding initial attenuation threshold from the threshold set and mark it as the current attenuation threshold; Extract the historical attenuation set, and perform mean processing on the historical attenuation in the historical attenuation set to obtain the historical average attenuation. If the historical average attenuation is greater than or equal to the current attenuation threshold, then the current attenuation threshold is used as the actual attenuation threshold. If the historical average attenuation is less than the current attenuation threshold, the difference between the current attenuation threshold and the historical average attenuation is calculated. The current attenuation threshold is then subtracted from the product of the difference and a preset smoothing coefficient to obtain the corrected current attenuation threshold. The corrected current attenuation threshold is then used as the actual attenuation threshold. The actual attenuation threshold is used as the preset attenuation threshold. The regular attenuation degree is compared with the preset attenuation threshold. If the regular attenuation degree is less than the preset attenuation threshold, the pricing topology reconstruction instruction is not generated. If the regular attenuation degree is greater than or equal to the preset attenuation threshold, the pricing topology reconstruction instruction is generated.

[0010] Optionally, methods for generating pricing compensation parameters and calculation logic paths include: Calculate the difference between the regular attenuation degree and the actual attenuation threshold, and use the ratio of the difference to the actual attenuation threshold as a compensation coefficient; The pricing compensation parameter is obtained by multiplying the compensation coefficient by the preset benchmark pricing parameter. Obtain all basic logic nodes in the predefined audit compliance logic; A directed acyclic graph is constructed using the basic logical nodes as vertices and the logical dependencies between nodes as edges. The weights of the edges are updated using the pricing compensation parameters. The shortest directed connected path that satisfies the preset audit constraints is searched in the directed acyclic graph with updated weights, in conjunction with the target reconstruction strategy. The shortest directed connected path is used as the computational logical path.

[0011] Optionally, methods for obtaining the dynamic compensation interpretability index include: Extract all newly added and modified logical nodes from the computational logical path to form a set of changed nodes; Perform a reverse mapping analysis between each node in the set of variable nodes and the standard cost quota item to obtain the mapping success identifier corresponding to each node; The mapping success rate is obtained by counting the number of successful mapping identifiers and dividing it by the total number of nodes in the changed node set. The mapping success rate is used as the interpretability index of the dynamic compensation.

[0012] Optionally, methods for identifying the evolutionary state of a pricing model based on a dynamic compensation interpretability index include: Obtain all the dynamic compensation interpretability indices evaluated during the system reconfiguration process under the current technical upgrade project type, and combine them into an interpretability index set; Generate an index change sequence based on the dynamically compensated interpretability index in the interpretability index set; Sliding window regression slope analysis is performed sequentially on the exponential change sequence to obtain the changing trend of the exponential change sequence within each sliding window; The sliding window with a negative trend is designated as the first analysis window; The pricing topology corresponding to the first analysis window is marked as a high-risk evolutionary state; The sliding window whose change trend is positive or zero is marked as the second analysis window; The pricing topology corresponding to the second analysis window is marked as a safe evolution state.

[0013] Optionally, after identifying the evolutionary state of the pricing model, the following may also be included: Obtain the pricing topology in the security evolution state and extract the core pricing logic structure therein; Obtain the feature vectors of other types of technological upgrading projects that are not involved in the current pricing calculation as the target heterogeneous feature vectors. The target heterogeneous feature vectors have a distribution difference from the benchmark feature space corresponding to the current technological upgrading project. The core pricing logic structure is used as a pre-trained model and transferred to the pricing calculation corresponding to the target heterogeneous feature vector for fine-tuning and testing. The number of nodes in the core pricing logic structure that can be directly reused in the fine-tuning test is counted, and then divided by the total number of nodes in the core pricing logic structure to obtain the cross-context feature generalization retention rate. The pre-constructed baseline feature space is updated based on the cross-context feature generalization retention rate.

[0014] The beneficial effects of this invention are: 1) This invention standardizes the format and timestamps the structural data of Building Information Modeling (BIM), IoT time-series data, and unstructured financial text, and constructs a cross-modal attention mechanism for feature splicing and fusion. This eliminates the misalignment of multi-source heterogeneous data in terms of time granularity and expression form, realizes the joint representation of specific construction stages, and effectively solves the problem of inconsistent data formats. 2) This invention performs sliding window segmentation, centering, and whitening on the basic feature representation, and calculates the feature distribution offset through principal component analysis and maximum mean difference; further, it uses dynamic time warping algorithm to match the standard offset vector and quantifies the rule decay degree; this realizes operable quantitative monitoring of the degree of decline in environmental changes and rule applicability, and overcomes the serious lag of traditional static pricing. 3) This invention dynamically adjusts the attenuation threshold by combining the current project type with the historical average attenuation. When the rule attenuation exceeds the limit, a reconstruction instruction is triggered. The shortest directed connected path is searched in the directed acyclic graph by combining audit constraints, and pricing compensation parameters are generated. This enables the pricing topology to evolve adaptively with the site conditions, effectively overcoming the problem of fixed paths deviating from reality, and ensuring accurate cost calculation and logical traceability. 4) This invention reverse maps the variable nodes to the standard cost quota items to obtain the dynamic compensation interpretability index, and identifies the model evolution state through regression slope analysis. This not only evaluates the audit compliance of the pricing logic in a quantitative way and realizes risk warning, but also extracts the core pricing logic under the safe evolution state for cross-scenario migration, thereby improving the system's long-term adaptive capability in the face of complex technical transformation projects. Attached Figure Description

[0015] The invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a schematic diagram of the modules of the power grid technical transformation pricing system based on industrial data analysis provided in the embodiments of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 The power grid technical upgrade pricing system based on industrial data analysis includes: an industrial data acquisition module, used to acquire the current technical upgrade project type and timestamp information, and to acquire raw industrial data from multi-source heterogeneous data, as well as the current pricing rules; The data modality alignment module is used to perform cross-modal feature alignment and joint representation processing on the raw industrial data, and extract basic feature representations. The concept drift perception module is used to monitor the distribution of the basic feature representation by combining a preset historical benchmark to obtain the feature distribution offset, identify the concept drift state, and quantify the rule decay degree of the current pricing rule. The topology reconfiguration decision module is used to retrieve historical pricing topology structures and predefined audit compliance logic to determine the target reconfiguration strategy, compare the rule decay degree with the preset decay threshold, and determine whether to generate a pricing topology reconfiguration instruction. The dynamic compensation generation module is used to, if the pricing topology reconstruction instruction is not generated, use the historical pricing topology as the updated pricing topology and set default pricing compensation parameters; if the pricing topology reconstruction instruction is generated, combine the target reconstruction strategy to generate pricing compensation parameters and calculation logic path, and update the pricing topology to obtain the updated pricing topology. An adaptive evolution module is used to evaluate the updated pricing topology to obtain a dynamic compensation interpretability index and identify the evolution state of the pricing model. The pricing execution and output module performs pricing calculations on the original industrial data based on the updated pricing topology and the pricing compensation parameters, and outputs the final pricing results and cost list.

[0019] This embodiment provides a power grid technical upgrade pricing mechanism based on industrial data analysis; specifically, it focuses on the full-process cost calculation of a 500kV substation upgrade project in a coastal city, involving the replacement of old gas-insulated metal-enclosed switchgear and the expansion of the main transformer bay; the construction period of this project spans the rainy season and the concentrated equipment delivery period, and the on-site structural model, IoT monitoring and procurement financial documents are constantly changing, making it suitable as a dynamic pricing scenario. The details are as follows: The industrial data acquisition module reads the project initiation information, determines the current technical upgrade project type as a joint operation of GIS equipment upgrade and expansion, and simultaneously records a unified timestamp; this unified timestamp can be at the minute or hour level, for example, uniformly recorded as within a certain collection cycle. The system collects data from three channels: one is building information model structural data, such as interval dimensions, number of equipment foundations, and cable trench length; Secondly, there is the time-series data of IoT device status, such as the duration of old equipment removal, the load of hoisting equipment, and the temperature and humidity on site; thirdly, there is unstructured financial text, such as equipment quotations, transportation surcharge descriptions, and construction visa texts; after collection, a raw industrial data set is formed; the data modality alignment module performs joint representation on the above data; The structure of a building information model can be extracted as a vector:

[0020] Extracting IoT time series data into vectors:

[0021] And extract the financial text as a vector: ;

[0022] Then, vectors A, B, and c are transformed into the same feature space through a unified mapping to form the basic feature representation F; for example, after mapping, we can obtain:

[0023] This basic feature does not directly provide the cost, but rather serves as input for subsequent drift monitoring and rule-based judgment; The concept drift sensing module continuously monitors the distribution changes of the basic feature representation F in a continuous time window; Assume the mean values ​​of the basic features for the two time windows at the beginning of the project are respectively, and the mean value of the historical baseline features is respectively. The historical baseline characteristic mean is The system can then calculate the degree of offset; for example, the offset of window 1 is 0.06 and the offset of window 2 is 0.15. If multiple windows subsequently increase, it is identified as the change in the characteristic distribution of the pricing scenario exceeds the preset offset threshold, and the rule decay degree of the current pricing rule is further quantified; this rule decay degree can be understood as the decrease in the adaptability of the original quota logic to the current project scenario; for example, the decay degree obtained after quantification is 0.72. The topology reconstruction decision module retrieves the historical pricing topology from the pricing database. This topology can be abstracted into several logical nodes and their connections. For example, node N1 represents the calculation of demolition work volume, node N2 represents the application of main material prices, node N3 represents the adjustment of hoisting machinery shifts, node N4 represents the inclusion of ocean freight surcharges, and node N5 represents the verification of nighttime construction subsidies. The original connections can be represented as follows: ; At the same time, the system also reads audit compliance logic, such as new fees must be mappable to quota items or contractual agreements, duplicate billing paths must not coexist, and measures and fees must not be repeatedly added across professions; it compares the rule decay degree with the preset decay threshold; if the threshold is 0.65 and the decay degree is 0.72, it generates a pricing topology reconstruction instruction. The dynamic compensation generation module operates under two branches: if no reconstruction is triggered, it continues to use the historical pricing topology and sets the compensation parameters to preset default values, such as 1.00, indicating that only the usual quota is executed; if reconstruction is triggered, it generates pricing compensation parameters and a new calculation logic path in combination with the target reconstruction strategy. For example, due to changes in installation efficiency caused by coastal transportation surcharges and high-humidity operations, the system can initially obtain an incremental compensation parameter of 0.08. During actual pricing, to facilitate direct application to existing cost items, this parameter can be converted to an execution multiplier of 1.08, and the logical path can be updated accordingly. This leads to a new pricing topology. The incremental compensation parameter here describes the increase relative to the benchmark pricing parameter, while the execution multiplier is the business execution form after the increment is added to the benchmark value. The two correspond to the same round of reconstruction results, only the accounting methods are different. The adaptive evolution module performs interpretability assessments on the updated topology; for example, whether the newly added N4 ocean surcharge can be mapped to the contract supplementary clauses, and whether the revision of N3 can be mapped to the basis for adjusting machine shifts. If the mapping success rate is greater than or equal to the preset success rate threshold, the corresponding dynamic compensation interpretability index, such as 0.92, is output; then, the index is used to identify whether the current model is in a safe evolutionary state or a high-risk evolutionary state. The pricing execution and output module performs calculations on the original industrial data based on the updated pricing topology and compensation parameters. Taking the above scenario as an example, the system first calculates the demolition work volume, then applies the main equipment material price, then adds the ocean freight surcharge and hoisting efficiency correction, and outputs the final pricing result and cost list. The cost list may include equipment costs, installation costs, measures costs, transportation surcharges, and explanatory fields for subsequent auditing and review. As a supplementary explanation, the system does not directly stop when data for a certain modality is missing; if BIM data arrives late, the most recently verified structural version is used as a temporary input and marked as pending review; if the IoT timing signal is interrupted, the missing window interpolation or the average of adjacent windows is used as a substitute. If the financial text cannot be parsed, the corresponding expense node will be locked for manual review and then released without automatic compensation; if the rule decay rate is close to the threshold but not reached, for example, 0.64 close to 0.65, it can enter the observation state and be continuously tracked in one or more subsequent time windows to avoid frequent reconstruction. For example, in this 500kV substation project, during the middle of the project, due to the delay in the arrival of imported GIS bushings, the increase in ocean freight surcharges exceeded the preset cost change threshold, and at the same time, the hoisting efficiency index was lower than the preset efficiency lower limit. The system captures the joint offset of three types of data: BIM progress, IoT work load, and financial supplementary agreement. It recognizes that the original pricing logic of fixed main material price + standard machine shift can no longer fully reflect the on-site cost composition. Therefore, it automatically generates new compensation parameters and reconstructed logical paths, and retains the source basis of the new nodes when outputting the list. The purpose of this step is to enable the pricing system to no longer rely on static rules, but to complete the closed-loop processing of rule decay identification, topology reconstruction, interpretable evaluation and pricing output under the condition of continuous change of multi-source industrial data, so as to achieve dynamic, adaptive and auditable pricing calculation for power grid technical transformation projects.

[0024] In a preferred embodiment of the present invention, the method for acquiring raw industrial data includes: real-time acquisition of multi-channel business data, wherein the multi-channel business data includes building information model structure data, Internet of Things device status time-series data, and unstructured financial text data; Each type of multi-channel service data is subjected to format standardization processing to obtain standard format data corresponding to each type of multi-channel service data; based on the timestamp information, timestamp synchronization compensation is performed on the standard format data to obtain the original industrial data corresponding to each type of multi-channel service data.

[0025] This embodiment provides a raw industrial data acquisition step; specifically, in the aforementioned 500kV substation technical renovation project, although a multi-channel data access link has been established, if data from different sources is directly sent to subsequent modules, problems such as inconsistent time granularity, inconsistent field formats, and inconsistent data units will occur, leading to misalignment in subsequent joint representation; therefore, this embodiment introduces a format standardization and timestamp synchronization compensation process. The details are as follows: When the system obtains structural data from the BIM channel, it may read information such as component list, spatial coordinates, and professional codes; for example, in one data collection, it obtained 2 GIS intervals, 6 foundations, and a cable trench length of 180 meters; The IoT channel retrieves device status timelines, such as 08:00 hoisting load 0.72, 08:10 load 0.81, and 08:20 load 0.78; the financial text channel retrieves contracts, quotations, or visa statements, such as an additional 20,000 yuan per unit for ocean freight surcharges and night window operation fees, to be executed according to the supplementary terms of the contract; the original formats of these three types of data may be model files, timeline messages, and natural language text, respectively. When the format is standardized, BIM structural data can be uniformly converted into structured field tables, such as component type, quantity, size, discipline, and update time; IoT data can be uniformly converted into device number, sampling time, status value, and sampling quality mark; financial text can be converted into expense category, amount, trigger condition, valid range, and source document number through text parsing; for example, the ocean surcharge statement is parsed to form record R1=Expense Category: Transportation Surcharge, Amount: 20000, Trigger Condition: Imported Equipment, Document Number: D008; After format unification is completed, the system performs timestamp synchronization compensation. Assuming the BIM update occurs at 09:00, the IoT sampling cycle is once every 10 minutes, the financial text supplementary agreement is entered at 09:23, and the unified pricing time for this round is set at 09:30, the system can summarize multiple IoT samples between 09:00 and 09:30 into a window feature, such as an average load of 0.77 and a peak load of 0.83; the financial text entered at 09:23 is marked as valid for the current window. For BIM structures that have not been changed before 09:30, the current window structure state will be directly used; ultimately, a set of synchronized original industrial data will be formed at a unified time point. To demonstrate the synchronous compensation process, a specific calculation example is as follows: Assume the BIM data update time is... The original time series of IoT contains Three points, the financial text update time is The unified timestamp is The system maps BIM data to the nearest valid rule. Map IoT data to according to window aggregation rules Map financial documents to the system according to the rule that they take effect immediately upon entry. ,form Complete raw industrial data package at the current moment; Furthermore, if a channel does not have valid data before a unified timestamp, different processing will be adopted; if the BIM version is too old and exceeds the allowable time difference, such as not being updated for more than 48 hours, the structural data will be marked as expired and will not directly participate in automatic pricing. If the IoT has insufficient sampling points in the current window, for example, only 1 point, while the preset minimum sampling number is 3, then adjacent window compensation is enabled and the modality weight is reduced; if the financial text lacks an amount but has an expense category, then the category node is retained first, the amount field is set to empty and the process is transferred to manual confirmation; if there are multiple conflicting versions of the same expense clause, then the latest approved version shall prevail, while the historical version index is retained for audit backtracking. For example, during the centralized equipment delivery phase of the project, the BIM model confirmed the addition of cable brackets at 9:00 AM, the IoT acquisition system continuously recorded the load of the hoisting equipment from 9:00 AM to 9:30 AM, and the financial system entered the transportation supplementary agreement at 9:23 AM. After synchronizing the three data to a unified time point of 9:30 AM, the system obtained raw industrial data that could reflect changes in structure, workload, and cost terms, providing reliable input for subsequent drift sensing. The purpose of this step is to eliminate the misalignment in format and time between multi-channel heterogeneous data, thereby achieving input consistency for subsequent feature extraction and joint representation.

[0026] In a preferred embodiment of the present invention, the method for cross-modal feature alignment and joint characterization processing of the raw industrial data includes: extracting independent features for each type of raw industrial data to obtain independent modal feature vectors; obtaining the timestamp information and the current technical transformation project type, and obtaining a corresponding spatial mapping matrix from a pre-constructed benchmark feature space based on the timestamp information and the current technical transformation project type; the spatial mapping matrix characterizes the degree of mutual influence between different modalities of multi-source heterogeneous data; Based on the spatial mapping matrix, a cross-modal attention mechanism is constructed. The specific construction process of the cross-modal attention mechanism is as follows: each independent modal feature vector is multiplied by the spatial mapping matrix to obtain the initial attention score of each modality, and the initial attention score is processed by the Softmax normalization function to obtain the final assigned weight corresponding to each independent modal feature vector. The cross-modal attention mechanism is used to assign weights to each independent modal feature vector. Specifically, the initial attention score is processed by a normalization function to obtain the final assigned weights, and each weighted independent modal feature vector is spliced ​​and fused to complete the cross-modal feature alignment and joint representation processing of the original industrial data.

[0027] This embodiment provides a cross-modal feature alignment and joint representation step; specifically, simply completing data synchronization is not enough to stably support pricing judgment, because structural data, time-series data and text data have different ways of expression. If they are simply spliced ​​together, it is easy to cause one type of modality to be too strong or too weak, especially when the project type changes or the construction stage changes; therefore, this embodiment further introduces a spatial mapping and cross-modal weight allocation mechanism based on timestamps and project types. The details are as follows: three independent modal features are extracted; for BIM structural data, structural scale, component complexity, and interdisciplinary crossover can be extracted to obtain structural feature vectors. For IoT time-series data, the average load, load fluctuation, and power outage window tension can be extracted to obtain a time-series feature vector. ; For financial texts, signals of main material price increases, signals of non-routine expenses, and contract supplementation strength can be extracted to obtain text feature vectors. Here, X represents the financial text feature vector, used to distinguish it from the timestamp identifier mentioned earlier. and other texts below Distinguish between prefixed time variables; The system selects the corresponding spatial mapping matrix from the pre-constructed baseline feature space based on the unified timestamp and the current technical renovation project type; for example, if the project is in the GIS update + expansion stage and the rainy season construction stage, then the mapping matrix M is read. For ease of explanation, assume M is a The simplified matrix, whose first row is... The second line The third line This matrix represents the degree of mutual influence between different modalities at the current project stage. The value of 0.2 in the second row and third column indicates that, at the current stage, time-series information has a moderate impact on the features of financial text. A cross-modal attention mechanism is constructed based on this mapping matrix; the aforementioned cross-modal weight allocation mechanism is the specific implementation form of this cross-modal attention mechanism in this embodiment; S, I, and X can be interacted with matrix M respectively to obtain three weighted scores, for example... Then, normalize these three scores to obtain the weights applied to the structural feature vector S, the temporal feature vector I, and the textual feature vector X, respectively. ; The weighted feature vectors are concatenated and merged to form the basic feature representation. This vector simultaneously preserves information within each mode and the phased effects between modes. Here, a simplified deduction can be provided: if, in the early stages of the project, the change in financial text features is lower than the preset fluctuation threshold, then the weight corresponding to the text modality... It might only be 0.20; however, during the centralized equipment procurement phase, supplementary agreements to contracts frequently appear, increasing the weight corresponding to the text modality. It can rise to 0.36; thus, the joint representation of the same BIM structural change will be different at different stages, which is more in line with the actual cost scenario; Furthermore, if the data quality index of a certain modality in the current data window is lower than the preset quality threshold, it will not be forced to participate in the fusion at a fixed ratio; for example, if the IoT channel experiences a large area of ​​missing data due to network fluctuations on site, and the quality label is lower than the preset value of 0.6, the system will reduce its weight, or even temporarily remove it and re-normalize the other modalities. If there is no perfectly matching mapping matrix for the current technical upgrade project type in the benchmark feature space, the closest project type matrix is ​​used and a learning identifier is added. At the same time, the current sample is included in the subsequent benchmark space update set. If all three modalities are valid but contradict each other, such as BIM showing an increase in engineering volume while financial text reflects a decrease in costs, the conflict marker is retained for comprehensive judgment by the subsequent drift module, rather than forcibly resolving the conflict in this step. For example, after the 500kV substation project transitioned from the routine dismantling to the equipment installation phase, the system detected a dense increase in supplementary financial agreements, the IoT showed that the hoisting load was consistently high, and the BIM added temporary support structures. By redistributing weights through a mapping matrix based on the rainy season installation phase, the influence of text and temporal modalities on the final joint features is enhanced, enabling subsequent rule decay identification to detect earlier that the original standard installation unit price can no longer accurately cover the changing trend of current on-site resource consumption. The purpose of this step is to map industrial data from different sources and in different forms of expression into a unified and comparable feature space, thereby achieving joint representation for specific project types and specific construction stages.

[0028] In a preferred embodiment of the present invention, the method for obtaining the feature distribution offset includes: dividing the basic feature representation into sliding windows to obtain multiple current data windows; performing centering processing on each current data window to obtain a centered feature; and performing whitening processing on each centered feature to obtain a whitened feature. The whitening process specifically employs PCA whitening or ZCA whitening algorithms. By calculating the covariance matrix of the centered feature and obtaining its eigenvalues ​​and eigenma, the feature is projected onto a new coordinate system and divided by the square root of the eigenvalue, so that each dimension of the feature has the same variance and is uncorrelated with each other. Principal component analysis is performed on all the whitening features to determine the main feature dimensions; benchmark feature vectors corresponding to the main feature dimensions are extracted from the preset historical benchmark distribution and a benchmark matrix is ​​formed; the centered features of each current data window are combined to form the current feature matrix, the covariance matrix of the current feature matrix is ​​calculated, and the covariance matrix is ​​decomposed to obtain the current feature vector matrix; A Gaussian radial basis function is introduced to map the current eigenvector matrix and the reference matrix to the regenerating kernel Hilbert space. The maximum mean difference between the current eigenvector matrix and the reference matrix is ​​calculated, and the maximum mean difference is used as the feature distribution offset corresponding to each current data window.

[0029] This embodiment provides a step for obtaining feature distribution offset. Specifically, after completing the joint characterization, if only the feature value at a single point in time is observed, occasional fluctuations may be misjudged as rule failures. For example, a higher load on a single hoisting operation on a certain day does not necessarily mean that the pricing logic needs to be reconstructed. Therefore, this embodiment identifies stable distribution changes through sliding window, centering, whitening, and principal dimension extraction. The details are as follows; assuming the system continuously obtains basic feature representations at 6 time points, each time point is simplified and explained using only two dimensions: , , , , , If the window length is set to 3 and the step size to 1, then 4 current data windows can be formed: , , , ; Centralize the processing of each window; For example, its mean is After centralization, ; Further whitening can be understood as scaling different dimensions to a comparable state, preventing any one dimension from dominating the result; assuming the whitened... available , , Principal component analysis was performed on the whitening features of all windows to screen out the main feature dimensions. For example, the contribution rates of the first and second dimensions were 55% and 35% respectively, totaling 90%, so they were both retained as the main feature dimensions. Extract the corresponding dimension of the baseline feature vector from the historical baseline distribution to construct the baseline matrix. The historical baseline matrix is ​​denoted here as Using the letters that represent IoT modal vectors as mentioned above Distinguish between them; assume the historical baseline matrix is ; Then, the centered features of each current window are used to construct the current feature matrix, and its covariance matrix is ​​calculated; For example, if its covariance matrix is ​​decomposed into its eigenvector matrix, then... Then the matrix is ​​further calculated. and The maximum mean difference between them; assuming the difference is 0.31, then 0.31 is taken as... The corresponding feature distribution offset; For ease of understanding, a continuous window comparison can be performed; if , , and The offsets are 0.08, 0.14, 0.31, and 0.35 respectively, indicating that... The project feature distribution shows a significant change, and this change is not a single point of noise, but a continuous increase; this window-level offset is more suitable for triggering subsequent rule-based decay quantization than single-point comparison. Furthermore, if the number of samples in a window is insufficient, for example, only 2 points are obtained due to data acquisition interruption, and the preset window length is 3, then the window will not directly participate in the offset calculation, but can be merged with the adjacent window before processing; If the variance of a certain dimension is close to 0 after centering, the whitening process may be unstable. In this case, a minimum stable term can be introduced for that dimension to prevent abnormal amplification. If the cumulative contribution rate after principal component analysis is lower than the preset threshold, such as lower than 80%, it means that the current feature dispersion is greater than the preset dispersion threshold. The system can temporarily expand the window length and recalculate to reduce the impact of occasional factors. If the baseline distribution lacks samples for the corresponding project stage, the baseline distribution of the most recent stage of the same type of project will be used as a substitute, and the current shift result will be marked as low confidence. For example, in the coastal substation technical renovation project, the distribution of features of each window in the early stage of the project was similar to that of the historical GIS update project, with the offset remaining around 0.1; after entering the stage of equipment delivery delay and rainy season overlap, the offset rose continuously to over 0.3, reflecting that the combined changes in construction organization, equipment transportation and material prices have significantly deviated from the historical normal. The purpose of this step is to transform the subjective judgment of whether environmental changes have occurred into a calculable window-level distribution offset, thereby achieving quantitative monitoring of the applicability of pricing rules.

[0030] In a preferred embodiment of the present invention, the method for quantifying the rule decay of the current pricing rule includes: forming an offset time series by arranging multiple feature distribution offsets in chronological order; smoothing the offset time series to obtain a smoothed offset sequence; and using a dynamic time warping algorithm to calculate the degree of matching between the smoothed offset sequence and the standard offset vectors corresponding to different decay levels in a pre-established rule decay mapping library. The matching degree is calculated as follows: the cumulative path distance between the smoothed offset sequence and the standard offset vector is calculated by the dynamic time warping algorithm, and the cumulative path distance is back-mapped to a matching degree score between 0 and 1 using a negative exponential function or a preset normalization formula. The smaller the cumulative path distance, the higher the matching degree score. Extract the preset value of the attenuation level corresponding to the standard offset vector whose matching degree is greater than the preset matching threshold. The preset value represents the risk of failure of the pricing rule. The preset value is used as a candidate attenuation coefficient. If the matching degree of all standard offset vectors is less than or equal to the preset matching threshold, the current rule attenuation degree is recorded as the preset minimum default value. The largest candidate attenuation coefficient is used as the rule attenuation degree of the current pricing rule. The rule attenuation degree calculated each time is stored in the historical attenuation degree set.

[0031] This embodiment provides a step for measuring rule decay. Specifically, since the window offset only reflects the degree of change in data characteristics and cannot directly characterize the failure level of the original pricing rule, it is necessary to further perform decay mapping to assist in generating reconstruction instructions. Therefore, this embodiment matches the offset time series with a preset decay mode to obtain the rule decay degree. The details are as follows; assuming that the offset sequence of 5 consecutive windows is obtained... First, smooth the sequence, for example, by using the neighborhood mean method, to obtain a smoothed offset sequence. The purpose of smoothing is to reduce the impact of noise in individual windows. The system pre-establishes a rule-based attenuation mapping library; this library can store several standard offset vectors and their corresponding attenuation levels; for example, levels Corresponding standard vector Its default value is 0.2; grade Corresponding standard vector Its default value is 0.4; Level Corresponding standard vector Its default value is 0.7; level Its default value is 0.9; The system uses a dynamic time warping algorithm to calculate respectively and to The degree of matching between them; assuming the matching results are 0.42, 0.61, 0.89 and 0.73 respectively; when the preset matching threshold is 0.75, only L3 meets the requirements, so the candidate attenuation coefficient includes 0.7, and the final rule attenuation is recorded as 0.7; If multiple levels simultaneously meet the threshold, the largest candidate attenuation coefficient is taken as the current rule attenuation degree; for example, if and The matching degrees were 0.80 and 0.78, respectively, so the candidate attenuation coefficients were 0.7 and 0.9, respectively, and finally 0.9 was chosen; This method of taking the maximum value can preferentially reflect higher failure risks among multiple similar patterns; each obtained rule decay degree is written into the historical decay degree set to prepare for subsequent adaptive threshold adjustment; A specific calculation example is as follows: If the attenuation rates obtained for the first three days of this week are 0.42, 0.55, and 0.70 respectively, then the historical attenuation rate set will gradually accumulate to... This indicates that the applicability of pricing rules is declining continuously, rather than being a random fluctuation. Furthermore, if the length of the smoothed offset sequence is inconsistent with the length of the standard offset vector, the length can be unified by truncation or interpolation before matching; if the matching degree of all standard offset vectors is lower than the threshold, a high-risk conclusion is not directly output, but the current sequence is classified into the unknown decay mode, the nearest neighbor level is temporarily used as the reference value, and the sequence is sent to the mapping library expansion queue. If the candidate decay coefficient is empty, the current rule decay degree can be recorded as the minimum default value, such as 0.1, and continued to be observed in subsequent time windows; if the capacity of the historical decay degree set reaches the preset upper limit, the oldest data will be eliminated in chronological order to avoid interference from the current judgment due to too early state. For example, in this substation project, the pricing rules originally applicable to conventional GIS replacement projects showed a typical medium-to-high attenuation pattern in the offset sequence after several rounds of procurement supplementary agreements, temporary transportation adjustments, and high humidity construction conditions. After comparison with the mapping library, the system identified that the current rule attenuation reached 0.7, indicating that relying solely on the original rules was no longer sufficient to fully cover the actual cost composition. The purpose of this step is to convert continuous offsets into rule decay that can be used for decision-making, thereby enabling actionable quantification of the degree to which data changes lead to a decrease in rule effectiveness.

[0032] In a preferred embodiment of the present invention, the method for determining whether to generate a pricing topology reconfiguration instruction includes: a preset threshold set, wherein the threshold set includes initial attenuation thresholds corresponding to different technical renovation project types; and according to the current technical renovation project type, obtaining the corresponding initial attenuation threshold from the threshold set and marking it as the current attenuation threshold. Extract the historical attenuation set, average the historical attenuation values ​​in the historical attenuation set to obtain the historical average attenuation value; if the historical average attenuation value is greater than or equal to the current attenuation threshold, then the current attenuation threshold is used as the actual attenuation threshold; if the historical average attenuation value is less than the current attenuation threshold, then calculate the difference between the current attenuation threshold and the historical average attenuation value, subtract the product of the difference and a preset smoothing coefficient from the current attenuation threshold to obtain the corrected current attenuation threshold, and use the corrected current attenuation threshold as the actual attenuation threshold; The actual attenuation threshold is used as the preset attenuation threshold. The regular attenuation degree is compared with the preset attenuation threshold. If the regular attenuation degree is less than the preset attenuation threshold, the pricing topology reconstruction instruction is not generated. If the regular attenuation degree is greater than or equal to the preset attenuation threshold, the pricing topology reconstruction instruction is generated.

[0033] This embodiment provides a pricing topology refactoring instruction judgment step; specifically, after obtaining the rule decay degree in the previous embodiment, if a fixed threshold is directly used for judgment, two types of problems are likely to occur: first, the frequency of triggering refactoring instructions for some highly volatile project types is higher than the preset frequency threshold; second, the refactoring response time for project types that change rapidly lags behind the preset update cycle, thus missing the adjustment opportunity; therefore, this embodiment introduces an initial threshold related to the project type and makes corrections based on historical decay levels. The following is a detailed description; the system has a preset threshold set; for example, the initial threshold for a regular transformer maintenance project is 0.75, the initial threshold for a GIS equipment upgrade project is 0.65, and the initial threshold for a joint expansion operation project is 0.60; in this main scenario, the project type is a GIS equipment upgrade + joint expansion operation, so a stricter threshold can be selected according to the preset rules, and the current attenuation threshold is set to 0.60 or 0.65; for ease of explanation, let's assume it is 0.65; The system calculates the historical average attenuation from the historical attenuation set; if the historical attenuation set The historical average attenuation is 0.56. Since 0.56 is less than the current attenuation threshold of 0.65, 0.65 is not used directly; instead, a correction is performed. Let the difference be 0.09 and the smoothing coefficient be 0.5. Then the corrected current attenuation threshold is... This value serves as the actual attenuation threshold; if the current rule attenuation is 0.70, then due to... The system generates a pricing topology reconfiguration instruction; If the historical average decay rate is already higher than or equal to the current threshold, for example, reaching 0.68, while the current threshold is 0.65, then 0.65 will be used directly as the actual threshold and will not be lowered further. This means that such projects have entered a state of high volatility and the threshold should not be lowered further to prevent the system from starting reconstruction too frequently. This approach can be understood as taking into account both the basic sensitivity of the project type and the recent historical status, so that the threshold is more realistic; for long-term stable projects, the threshold can be appropriately adjusted to improve responsiveness; for projects that are inherently highly volatile, the original threshold is maintained to maintain decision stability. Furthermore, if the historical decay set is empty, for example, when the system is processing a certain type of project for the first time, the initial threshold is directly used as the actual threshold; if the corrected threshold is lower than the system's allowed lower limit, for example, lower than 0.40, it is automatically raised to the lower limit value to avoid triggering reconstruction due to slight fluctuations. If the rule decay is exactly equal to the actual threshold, it will be processed as a triggered reconstruction to avoid long-term suspension when it is at the boundary value; if the current project type has no exact match in the threshold set, the threshold of the closest professional category will be selected and the source of this replacement will be recorded; if the deviation of the historical average decay caused by abnormally high values ​​exceeds the preset tolerance range, the extreme value mean or median can be used to replace the ordinary mean. For example, after the coastal substation project entered the rainy season, although the offset of some windows increased significantly, the system did not simply use a fixed threshold of 0.65 to judge. Instead, it adjusted the threshold to 0.605 based on the history of the attenuation in recent rounds. After the current attenuation reached 0.70, the system determined that the original pricing topology should enter the reconstruction process instead of continuing to use the old logic. The purpose of this step is to make the decision on whether to refactor take into account the differences in project type and historical evolution, so as to achieve more stable and more practical trigger control.

[0034] In a preferred embodiment of the present invention, the method for generating pricing compensation parameters and calculation logic paths includes: calculating the difference between the rule attenuation degree and the actual attenuation threshold, and using the ratio of the difference to the actual attenuation threshold as a compensation coefficient; multiplying the compensation coefficient by a preset benchmark pricing parameter to obtain the pricing compensation parameters; and obtaining all basic logic nodes in the predefined audit compliance logic. A directed acyclic graph is constructed using the basic logical nodes as vertices and the logical dependencies between nodes as edges. The weights of the edges are updated using the pricing compensation parameters. In the directed acyclic graph with updated weights, a Dijkstra shortest path search algorithm with conditional constraints is used in conjunction with the target reconstruction strategy to search for the shortest directed connected path that satisfies the preset audit constraints. The shortest directed connected path is then used as the computational logical path.

[0035] This embodiment provides a step for generating pricing compensation parameters and calculation logic paths. Specifically, to solve the problem that the system cannot execute the complete pricing calculation process, it is necessary to further clarify the reconstructed compensation range and the reconstructed calculation logic path. Especially in the power grid cost scenario, new or changed costs must be unfolded through a clear logical chain and cannot be directly given an untraceable total price correction value. Therefore, this embodiment generates compensation parameters and a logic path that meets audit constraints simultaneously after the reconstruction is triggered. The details are as follows: First, calculate the compensation coefficient; assuming the current rule attenuation is 0.70 and the actual attenuation threshold is 0.605, the difference between the two is 0.095, and the ratio of the difference to the actual attenuation threshold is approximately 0.157; if the preset benchmark pricing parameter is 1.00, then according to the calculation method in this embodiment, the obtained pricing compensation parameter is 0.157. This parameter represents the incremental increase required relative to the baseline value. In actual business operations, to facilitate direct adjustments to expense items, the system can further convert this incremental increase into an execution multiplier, i.e., 1.00 + 0.157 = 1.157. If the baseline pricing parameter is set separately according to expense categories, for example, 1.00 for materials, 0.80 for machinery, and 0.60 for measures, more refined incremental compensation parameters can be obtained, such as 0.157 for materials, 0.126 for machinery, and 0.094 for measures; the corresponding execution multipliers would be 1.157, 0.926, and 0.694, respectively. Therefore, the pricing compensation parameters in the embodiment serve to update the graph weights and describe the incremental magnitude, while when pricing is executed, a multiplier of the base value and the incremental value can be used, and the two can correspond one-to-one within the system. Read the basic logic nodes in the audit compliance logic; taking this project as an example, it can be set as follows: For the purpose of determining the amount of demolition work, The unit price for the main equipment was determined. For the determination of transportation surcharges, Corrections for machine operation. For verification of the cost of the measures, To exclude duplicate billing, Verification of supplementary clauses in the contract. For the final summary output; Construct a directed acyclic graph using these nodes as vertices, where edges represent sequential dependencies, for example... , , , , , Initial edge weights can represent execution costs, audit risks, or path priorities. The edge weights are updated using pricing compensation parameters; if the correlation weight of a certain type of cost is greater than a preset correlation threshold, the edge weights of the relevant paths will change; for example, transportation surcharges and machine shift adjustments are the focus of this round of restructuring. , and The weights of the three edges are reduced, indicating that these nodes should be included in the new path. Suppose that there are two candidate paths in the updated graph, where path PA is The total weight is 6.2; the path PB is... The total weight is 5.4; If the target reconstruction strategy requires prioritizing the retention of complete contractual paths, and the audit constraints require that new expenses must be verified by the contract or quota clauses, then path PB, which includes N3 and N7, is more suitable and is therefore selected as the shortest directed connected path and used as the computational logic path. Here, path PA and path PB are used to represent candidate paths to distinguish them from the letters A and B that represent different data vectors in the previous implementation. In this way, the system can not only determine the range of compensation parameters, but also clarify the corresponding calculation logic chain; each node in the path has business significance, which facilitates review by pricing personnel and auditors. As a supplementary explanation, if the rule attenuation degree just exceeds the threshold and the obtained compensation coefficient is less than the preset effective compensation lower limit, for example, only 0.02, then the system can limit the minimum effective compensation range. If it is lower than this range, the path will be rearranged without adjusting the specific parameters. If no complete path that simultaneously meets the audit constraints can be found in the graph, the process will revert to the most recently approved historical path and only allow insertion of nodes that are clearly defined by the contract. If multiple candidate paths have the same total weight, the one with fewer new nodes will be selected first to reduce unnecessary structural changes. If the input that a node depends on is missing, such as the amount of transportation surcharge not yet confirmed, the node will be retained but locked as pending data entry. The path can first become a semi-closed state, and the final calculation will be performed after the data is complete. For example, in this project, the transportation method of imported GIS equipment was temporarily changed from land transportation to sea transportation plus short-haul intermodal transportation, and the hoisting efficiency decreased due to the rainy season. The system calculated that the compensation parameters for this round need to be tilted towards the two cost categories of transportation and machinery. In the logic diagram, the path of purchase unit price → transportation surcharge → contract supplementary clause verification → machinery shift correction → measure cost verification is selected to ensure that each new correction has a clear source link. The purpose of this step is to transform the abstract changes brought about by rule decay into specific executable compensation parameters and traceable computational logic paths, thereby achieving controllable reconstruction of the pricing topology.

[0036] In a preferred embodiment of the present invention, the method for obtaining the dynamic compensation interpretability index includes: extracting all newly added logical nodes and modified logical nodes in the calculation logical path to form a set of changed nodes; performing reverse mapping analysis between each node in the set of changed nodes and the standard cost quota item to obtain the mapping success identifier corresponding to each node; The number of successfully mapped identifiers is counted and divided by the total number of nodes in the set of changed nodes to obtain the mapping success rate; the mapping success rate is used as the dynamic compensation interpretability index.

[0037] This embodiment provides a step for obtaining the interpretability index of dynamic compensation. Specifically, although a new compensation path was obtained in the previous embodiment, if the added or modified nodes cannot be mapped to standard quotas, contract terms, or audit basis, it is still difficult to meet the audit requirements of power grid engineering settlement scenarios. Therefore, this embodiment measures the interpretability of path changes through reverse mapping. The details are as follows: The system extracts all newly added and modified logical nodes from the computational logical path generated in this round; for example, in path B, compared to the historical path, N3 transportation surcharge determination and N7 contract supplementary clause verification have been added, while the parameter range of N4 machine shift correction has been modified. Therefore, the set of changed nodes can be denoted as... ; Perform reverse mapping analysis on each node in set U; the reverse mapping here is performed using the standard cost quota item as the main index; If the direct basis of a node comes from a supplementary contract agreement or approved document, it is first merged into the corresponding standard cost quota main item through the pre-maintained clause-quota linkage table in the system, and then it is determined whether the mapping is successful; that is, the contract clauses and approved documents are used as auxiliary evidence chains for quota items in this step, rather than being counted separately from the quota main index. Assuming the mapping results are as follows: N3 can be mapped to the standard cost quota item corresponding to secondary transportation and special logistics costs of equipment via the attachment table, and the success indicator is 1; N7 can be mapped to the main item of the standard cost quota corresponding to the verification of additional costs stipulated in the contract via the attachment table, and the success indicator is 1; if N4 can only be partially mapped to the machine shift quota but lacks the correction basis for high humidity operation, it is temporarily recorded as 0; then the number of successful mapping indicators is 2, the total number of change nodes is 3, and the interpretability index is... That is, 0.667; If an approved high-humidity operation correction document is subsequently added, and N4 is also mapped, the interpretability index will increase to 1.0. This shows that the index not only assesses the current state, but also reflects the change in interpretability after the data is improved. Furthermore, if the set of changed nodes is empty, it means that no new or modified logical nodes were added in this round. In this case, the interpretability index can be directly recorded as 1.0, indicating that the existing interpretable path was used. If a node can correspond to multiple quota items or multiple contract terms, the one with the shortest evidence chain and the clearest amount attribution is selected as the main mapping, and the rest are archived as auxiliary mappings. If a node cannot be directly mapped but can be interpreted by combining two or more approved nodes, it can be set as a successful combination mapping to avoid over-conservatism. If an abnormal denominator occurs when the total number of changing nodes is not 0, the system will directly stop the current round of exponent calculation and issue a structure verification alarm to prevent invalid results from entering the next stage. For example, in this project, the new round of restructuring added the ocean freight surcharge node and the contract verification node, and adjusted the hoisting machinery shift node; the ocean freight surcharge can be linked to the corresponding transportation quota main item through supplementary agreements, the contract verification node can be linked to the filed surcharge verification main item, and the machinery shift correction only has partial interpretability at first due to the lack of approval for high humidity working conditions; based on this, the system gives a dynamic compensation interpretability index of 0.667 and prompts cost estimators to supplement the supporting materials as soon as possible; The purpose of this step is to evaluate, in a quantitative way, whether the newly added pricing logic has sufficient business and audit basis, so as to achieve real-time constraints on the auditability of the reconstruction results.

[0038] In a preferred embodiment of the present invention, the method for identifying the evolution state of the pricing model based on the dynamic compensation interpretability index includes: obtaining all the dynamic compensation interpretability indices evaluated during the system reconstruction process under the current technical transformation project type, and combining them into an interpretability index set; Based on the dynamically compensated interpretability indices in the interpretability index set, an index change sequence is generated; a sliding window regression slope analysis is performed on the index change sequence in sequence to obtain the change trend of the index change sequence within each sliding window; The sliding window with a negative trend is marked as the first analysis window; the pricing topology corresponding to the first analysis window is marked as a high-risk evolution state; the sliding window with a positive or zero trend is marked as the second analysis window; the pricing topology corresponding to the second analysis window is marked as a safe evolution state.

[0039] This embodiment provides a step for identifying the evolutionary state of a pricing model. Specifically, a single interpretability index cannot accurately assess the stability of the model's evolution. A single high index may only be due to the accidental completeness of local data. If the index declines for several consecutive rounds, it indicates that although the model is adapting to the new situation, its explanatory chain is weakening, posing an audit risk. Therefore, this embodiment identifies safe or high-risk evolutionary states by analyzing the trend of the index sequence. The following is a detailed description: The system collects the dynamic compensation interpretability index of multiple rounds of reconstruction under the same project type, forming an interpretability index set; Assuming that in this type of GIS update project, the index set is obtained through five consecutive rounds of reconstruction. Here, J represents the interpretability index set, to distinguish it from the symbol E representing the current eigenvector matrix in the previous implementation; an exponential change sequence is formed accordingly. A sliding window regression slope analysis was performed on this sequence; if the window length is set to 3, three windows can be formed: , , ; Calculate the trends of change for the three windows separately; for ease of explanation, assume the regression slopes are respectively , and Since all three values ​​are negative, all three windows are marked as the first analysis window, and their corresponding pricing topology is marked as a high-risk evolution state. Here's a contrasting example: If the index for a project in the subsequent three rounds is... If the window slope is positive, it means that although there are many new logics, their quota mapping and contract basis are constantly being improved, and the corresponding topology can be marked as a safe evolution state. Through this trend identification, the system no longer relies solely on whether a particular round is compliant, but examines whether the explanatory power is improving or deteriorating; this is especially crucial for long-term pricing models. Furthermore, if the length of the index set is less than the minimum analysis window length, trend judgment is temporarily suspended, and the current state is marked as under observation; if the slope of a window is very close to 0, for example, the absolute value is less than 0.005, it is treated as a zero value and classified into a safe evolution state to avoid false alarms caused by small fluctuations; if the index sequence contains abnormal null values, they are first filled with the average of the neighboring rounds; if there are too many consecutive null values, the evolution state identification is suspended. If the index fluctuates violently within the same window, and the simple slope is insufficient to express the risk, a fluctuation amplitude check can be added. If the amplitude exceeds the preset value, the window will be marked separately as requiring manual review. For example, in this coastal 500kV substation project, as temporary cost nodes are continuously added, although the system can continue to complete the pricing, if it is difficult to map the newly added nodes to a stable quota or contract basis for several consecutive rounds, the interpretability index will gradually decline from 0.92 to 0.74. The system thus determines that the current pricing topology is in a high-risk evolutionary state and prompts that further automatic expansion needs to be controlled, prioritizing the supplementation of the basis chain and specification mapping; The purpose of this step is to identify whether the evolution of the pricing model is healthy from the perspective of time continuity, thereby achieving risk warning for the automatic reconfiguration process.

[0040] In a preferred embodiment of the present invention, after identifying the evolution state of the pricing model, the method further includes: obtaining the pricing topology in the safe evolution state and extracting the core pricing logic structure therein; obtaining the feature vectors of other types of technical transformation projects that have not participated in the current pricing calculation as target heterogeneous feature vectors, wherein the target heterogeneous feature vectors have a distribution difference from the benchmark feature space corresponding to the current technical transformation project; The core pricing logic structure is used as a pre-trained model and transferred to the pricing calculation corresponding to the target heterogeneous feature vector for fine-tuning testing. The number of nodes of the core pricing logic structure that can be directly reused in the fine-tuning test is counted and divided by the total number of nodes of the core pricing logic structure to obtain the cross-context feature generalization retention rate. The pre-constructed benchmark feature space is updated based on the cross-context feature generalization retention rate.

[0041] This embodiment provides a cross-context migration and baseline feature space update step; specifically, in the previous embodiment, the system was already able to distinguish between safe evolution states and high-risk evolution states; However, if the mature logic obtained from security evolution is only applicable to the current project and cannot be migrated to other types of technical upgrade projects, the system will still need to re-establish the corresponding mapping relationship when facing new projects, and the adaptation process will be lengthy. Therefore, this embodiment extracts the reusable core logic after identifying the security evolution state and performs migration and fine-tuning in heterogeneous projects. The details are as follows: First, the core pricing logic structure is extracted from the pricing topology corresponding to the safety evolution state. Taking this main project as an example, after multiple rounds of verification, the system confirms that the following nodes are stable and fully explained: K1 structural engineering quantity verification, K2 main material price source verification, K3 contract supplementary clause verification, K4 duplicate billing exclusion, and K5 summary output verification. These 5 nodes and their connection relationships constitute the core pricing logic structure. The system reads the feature vectors of other heterogeneous projects that are not involved in the current pricing calculation; for example, it selects the 220kV transformer cooling system technical renovation project as the target project. This project differs from the GIS update project in terms of equipment type, cost structure, and operation method. Its target heterogeneous feature vector can be simplified as follows: Since the feature distribution corresponding to vector G is different from the current benchmark feature space, the original model cannot be directly applied. Instead, the core logical structure is used as the basis for pre-training and fine-tuning. During fine-tuning testing, the system attempts to reuse K1 to K5 in the target project. Assuming the test results are: K1 can be reused directly, K2 can be reused directly because the main material price acquisition method is still applicable, K3 can be reused directly because there are also supplementary contract clauses for this project, K4 can be reused directly, while K5 needs to be slightly adjusted due to differences in equipment expertise before it can be used. If we follow the direct reuse standard, the number of nodes that can be directly reused is 4, and the total number of nodes is 5. Therefore, the cross-context feature generalization retention rate is 0.8. The system then updates the baseline feature space based on the retention rate. If the retention rate is high, for example, above 0.8, it indicates that some logical abstractions in the original baseline space have strong generalization ability. The system can increase the priority of this type of core logic in cross-project mapping and include the feature samples of the target project in the baseline space extension area. If the retention rate is low, for example, only 0.3, it indicates that the two types of projects are quite different. Only a few common verification nodes are retained, while the remaining feature mappings need to be rebuilt into a new subspace that is more suitable. A simplified deduction can be made: assuming that the baseline feature space originally includes the GIS update project subspace and the expansion joint operation subspace, when the transformer cooling system technical renovation project obtains a generalization retention rate of 0.8 after migration testing, the system can add a general logic sublayer for electromechanical renovation on the basis of the original space, and float high-reusability nodes such as K1, K3 and K4 as cross-professional common structures. As a supplementary explanation, if the number of selected security evolution topologies is too large and they differ greatly from each other, the core structure with the highest frequency of occurrence and a consistently stable interpretability index will be extracted first; if the feature distance between the target heterogeneous project feature and the existing benchmark space is greater than the preset distribution difference threshold, resulting in the inability to form an effective path during the fine-tuning process, the migration will be stopped, only the general verification node will be retained and a new independent subspace will be created. If the cross-context feature generalization retention rate is in the critical range, such as between 0.45 and 0.55, the main baseline space will not be updated directly. Instead, it will first enter the grayscale validation area and be formally included after more projects have validated it. If some nodes cannot be directly reused, but only the cost mapping table needs to be replaced for adaptation, they can be recorded as quasi-reusable nodes and retained as weak connection structures during subsequent space updates. For example, after the GIS update project of the coastal substation was completed and put into stable operation, the system migrated the core logic with a clear audit chain and high reusability to the transformer cooling system technical renovation project in another city. Tests revealed that the nodes for verifying the quantity of work, supplementing and verifying the contract, and eliminating duplicate charges can be directly reused, with only the node for equipment-specific costs requiring local adjustments. Based on this, the system increases the weight of cross-project common logic in the baseline feature space, thereby enabling a more stable pricing path to be formed more quickly when facing new types of technical upgrade projects. The purpose of this step is to precipitate the verified and secure pricing evolution results into a transferable core structure, and continuously update the benchmark feature space accordingly, so as to achieve the long-term adaptive capability of the system under different power grid technical transformation scenarios.

[0042] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A power grid technical upgrade pricing system based on industrial data analysis, characterized in that, include: The industrial data acquisition module is used to obtain the current technical transformation project type and timestamp information, collect multi-source heterogeneous data to obtain raw industrial data, and the current pricing rules; The data modality alignment module is used to perform cross-modal feature alignment and joint representation processing on the raw industrial data, and extract basic feature representations. The concept drift perception module is used to monitor the distribution of the basic feature representation by combining a preset historical benchmark to obtain the feature distribution offset, identify the concept drift state, and quantify the rule decay degree of the current pricing rule. The topology reconfiguration decision module is used to retrieve historical pricing topology structures and predefined audit compliance logic to determine the target reconfiguration strategy, compare the rule decay degree with the preset decay threshold, and determine whether to generate a pricing topology reconfiguration instruction. The dynamic compensation generation module is used to, if the pricing topology reconstruction instruction is not generated, use the historical pricing topology as the updated pricing topology and set default pricing compensation parameters; if the pricing topology reconstruction instruction is generated, combine the target reconstruction strategy to generate pricing compensation parameters and calculation logic path, and update the pricing topology to obtain the updated pricing topology. An adaptive evolution module is used to evaluate the updated pricing topology to obtain a dynamic compensation interpretability index and identify the evolution state of the pricing model. The pricing execution and output module performs pricing calculations on the original industrial data based on the updated pricing topology and the pricing compensation parameters, and outputs the final pricing results and cost list.

2. The power grid technical upgrade pricing system based on industrial data analysis according to claim 1, characterized in that, Methods for obtaining raw industrial data include: Real-time acquisition of multi-channel business data, including building information model structure data, IoT device status time-series data, and unstructured financial text data; Each type of multi-channel service data is subjected to format standardization processing to obtain standard format data corresponding to each type of multi-channel service data; Based on the timestamp information described in claim 1, timestamp synchronization compensation is performed on the standard format data to obtain the original industrial data corresponding to each type of multi-channel business data.

3. The power grid technical upgrade pricing system based on industrial data analysis according to claim 2, characterized in that, The method for cross-modal feature alignment and joint characterization processing of the raw industrial data includes: Independent feature extraction is performed on each type of raw industrial data to obtain an independent modal feature vector; Obtain the timestamp information as described in claim 1 and the current technical renovation project type, and based on the timestamp information and the current technical renovation project type, obtain the corresponding spatial mapping matrix from the pre-constructed benchmark feature space; the spatial mapping matrix characterizes the degree of mutual influence between different modalities of multi-source heterogeneous data; Based on the aforementioned spatial mapping matrix, a cross-modal attention mechanism is constructed; The cross-modal attention mechanism is used to assign weights to each independent modal feature vector. Specifically, the initial attention score is processed by a normalization function to obtain the final assigned weights, and each weighted independent modal feature vector is spliced ​​and fused to complete the cross-modal feature alignment and joint representation processing of the original industrial data.

4. The power grid technical upgrade pricing system based on industrial data analysis according to claim 3, characterized in that, Methods for obtaining feature distribution offsets include: The basic feature representation is divided into multiple current data windows by a sliding window. Each of the current data windows is centralized to obtain centralization features; Each of the centered features is whitened to obtain a whitened feature; Principal component analysis was performed on all the whitening features to determine the main feature dimensions; From the preset historical baseline distribution, extract the baseline feature vectors corresponding to the main feature dimensions and construct the baseline matrix; The centered features of each current data window are combined to form a current feature matrix. The covariance matrix of the current feature matrix is ​​calculated. The covariance matrix is ​​decomposed to obtain the current feature vector matrix. A Gaussian radial basis function is introduced to map the current eigenvector matrix and the reference matrix to the regenerating kernel Hilbert space. The maximum mean difference between the current eigenvector matrix and the reference matrix is ​​calculated, and the maximum mean difference is used as the feature distribution offset corresponding to each current data window.

5. The power grid technical upgrade pricing system based on industrial data analysis according to claim 4, characterized in that, Methods for quantifying the rule decay of the current pricing rule include: The multiple feature distribution offsets are arranged in chronological order to form an offset time series, and the offset time series is smoothed to obtain a smooth offset sequence. The dynamic time warping algorithm is used to calculate the degree of matching between the smoothed offset sequence and the standard offset vectors corresponding to different attenuation levels in the pre-established regular attenuation mapping library; Extract the preset value of the attenuation level corresponding to the standard offset vector whose matching degree is greater than the preset matching threshold. The preset value represents the risk of failure of the pricing rule. The preset value is used as a candidate attenuation coefficient. If the matching degree of all standard offset vectors is less than or equal to the preset matching threshold, the current rule attenuation degree is recorded as the preset minimum default value. The largest candidate attenuation coefficient is used as the rule attenuation degree of the current pricing rule. The rule attenuation degree calculated each time is stored in the historical attenuation degree set.

6. The power grid technical upgrade pricing system based on industrial data analysis according to claim 5, characterized in that, Methods for determining whether a pricing topology reconfiguration instruction has been generated include: A preset threshold set, which includes initial attenuation thresholds corresponding to different types of technical renovation projects; According to the current technical renovation project type as described in claim 1, the corresponding initial attenuation threshold is obtained from the threshold set and marked as the current attenuation threshold; Extract the historical attenuation set, and perform mean processing on the historical attenuation in the historical attenuation set to obtain the historical average attenuation. If the historical average attenuation is greater than or equal to the current attenuation threshold, then the current attenuation threshold is used as the actual attenuation threshold. If the historical average attenuation is less than the current attenuation threshold, the difference between the current attenuation threshold and the historical average attenuation is calculated. The current attenuation threshold is then subtracted from the product of the difference and a preset smoothing coefficient to obtain the corrected current attenuation threshold. The corrected current attenuation threshold is then used as the actual attenuation threshold. The actual attenuation threshold is used as the preset attenuation threshold in claim 1. The regular attenuation degree is compared with the preset attenuation threshold. If the regular attenuation degree is less than the preset attenuation threshold, the pricing topology reconstruction instruction is not generated. If the regular attenuation degree is greater than or equal to the preset attenuation threshold, the pricing topology reconstruction instruction is generated.

7. The power grid technical upgrade pricing system based on industrial data analysis according to claim 6, characterized in that, The methods for generating pricing compensation parameters and calculation logic paths include: Calculate the difference between the regular attenuation degree and the actual attenuation threshold, and use the ratio of the difference to the actual attenuation threshold as a compensation coefficient; The pricing compensation parameter is obtained by multiplying the compensation coefficient by the preset benchmark pricing parameter. Obtain all basic logic nodes in the predefined audit compliance logic described in claim 1; A directed acyclic graph is constructed using the basic logical nodes as vertices and the logical dependencies between nodes as edges. The weights of the edges are updated using the pricing compensation parameters. In the directed acyclic graph with updated weights, the target reconstruction strategy described in claim 1 is combined with the Dijkstra shortest path search algorithm with conditional constraints to search for the shortest directed connected path that satisfies the preset audit constraints. The shortest directed connected path is used as the computational logical path.

8. The power grid technical upgrade pricing system based on industrial data analysis according to claim 7, characterized in that, Methods for obtaining the interpretability index of dynamic compensation include: Extract all newly added and modified logical nodes from the computational logical path to form a set of changed nodes; Perform a reverse mapping analysis between each node in the set of variable nodes and the standard cost quota item to obtain the mapping success identifier corresponding to each node; The mapping success rate is obtained by counting the number of successful mapping identifiers and dividing it by the total number of nodes in the changed node set. The mapping success rate is used as the interpretability index of the dynamic compensation.

9. The power grid technical upgrade pricing system based on industrial data analysis according to claim 8, characterized in that, Methods for identifying the evolutionary state of a pricing model based on a dynamic compensation interpretability index include: Obtain all the dynamic compensation interpretability indices evaluated during the system reconfiguration process under the current technical upgrade project type, and combine them into an interpretability index set; Generate an index change sequence based on the dynamically compensated interpretability index in the interpretability index set; Sliding window regression slope analysis is performed sequentially on the exponential change sequence to obtain the changing trend of the exponential change sequence within each sliding window; The sliding window with a negative trend is designated as the first analysis window; The pricing topology corresponding to the first analysis window is marked as a high-risk evolutionary state; The sliding window whose change trend is positive or zero is marked as the second analysis window; The pricing topology corresponding to the second analysis window is marked as a safe evolution state.

10. The power grid technical upgrade pricing system based on industrial data analysis according to claim 9, characterized in that, After identifying the evolutionary state of the pricing model, the following is also included: Obtain the pricing topology in the security evolution state and extract the core pricing logic structure therein; Obtain the feature vectors of other types of technological upgrading projects that are not involved in the current pricing calculation as the target heterogeneous feature vectors. The target heterogeneous feature vectors have a distribution difference from the benchmark feature space corresponding to the current technological upgrading project. The core pricing logic structure is used as a pre-trained model and transferred to the pricing calculation corresponding to the target heterogeneous feature vector for fine-tuning and testing. The number of nodes in the core pricing logic structure that can be directly reused in the fine-tuning test is counted, and then divided by the total number of nodes in the core pricing logic structure to obtain the cross-context feature generalization retention rate. The pre-constructed baseline feature space described in claim 3 is updated based on the cross-context feature generalization retention rate.

Images